Coordination toolkit for Claude Code. Cal manages the pipeline, not the code.
Cal is a coordinator, not a coder. It:
- Runs ideas through a dynamic pipeline
- Dispatches work to specialized agents
- Tracks work on GitHub Project boards (Epics + Features)
- Captures learnings
- Detects drift, frustration, and wrong assumptions
| Skill | Purpose |
|---|---|
/cal:next |
Advance pipeline — find and execute next step |
/cal:meet |
Meeting facilitator |
/cal:save |
Context preservation |
/cal:onboard |
Project setup + CLAUDE.md generation |
/cal:analyze [mode] |
Deep investigation (7 modes) |
/cal:cal-ood |
OOD principles reference |
| Mode | Best For |
|---|---|
inside-out |
Comprehensive understanding |
cake-walk |
Layering bugs (CSS, SwiftUI) |
rubberneck |
Focused scan with a suspect |
burst |
Temporal comparison |
bisect |
Binary search for root cause |
trace |
Follow data end-to-end |
diff-audit |
Catalog state differences |
These load every session via .claude/rules/:
| Rule | Detects |
|---|---|
coordinator |
Implementation requests → dispatches to agents |
tone-awareness |
Frustration or joy → adjusts approach |
squirrel |
Task drift → pauses and asks |
delta |
Wrong assumptions → documents correction |
.claude/rules/ → Behavioral rules (auto-loaded)
.claude/agents/ → Agent definitions (coder, reviewer, architect)
cal/ → Brain (learnings, state, preferences)
skills/ → Skill definitions (cal-ood, cal-analyze, cal-meet, etc.)
scripts/ → Shared utilities (gh-board.sh)
hooks/ → Plugin-level hooks (OOD enforcement)
ideas/ → Parking lot (unstructured)
docs/specs/ → Active work artifacts
Advancement requires explicit approval:
- "approved", "advance", "next phase", or
/approve - Positive feedback is not advancement